What is MatchEngine?

MatchEngine is a reverse image search engine that finds similar or duplicate versions of a given image. Besides duplicates, MatchEngine can find versions of the same image that have been altered in these ways:

  • cropped
  • flipped
  • warped or blurred
  • slightly rotated
  • color adjusted (including conversion to black-and-white)
  • watermarked or captioned

MatchEngine uses robust image fingerprinting and search algorithms, and is capable of performing very fast and accurate searches against large image sets. In the example below, the image on the left is the query image and the images on the right are the matches found, with relevance scores included:

_images/dedup_diagram.jpg

How does it work?

MatchEngine works by creating and comparing digital fingerprints of images. These unique digital fingerprints describe the patterns found in the pixels of the images. MatchEngine is able to identify even partial matches of these fingerprints.

MatchEngine creates an index of all your images’ fingerprints for reference and stores them in an internal database. API calls are used to add images to the index and to search through existing images.

Note that MatchEngine cannot extract fingerprints from images with no content (like solid color images) or from very small images. An error message will be returned if an attempt is made to index or search for an image that cannot be fingerprinted.

For MatchEngine to recognize an image it is searching for, there must be some identical portions of the images that match. MatchEngine will not recognize:

  • different photos of the same people (i.e., face recognition);
  • different photos of the same 3D object;
  • UPC or QR codes.